Title: Hostname validation in SSL match_hostname()
Type: enhancement Stage: resolved
Components: SSL Versions:
Status: closed Resolution: wont fix
Dependencies: Superseder:
Assigned To: christian.heimes Nosy List: Alex Gaynor, alex, christian.heimes, dstufft, janssen, ssivakorn
Priority: normal Keywords:

Created on 2017-03-16 08:03 by ssivakorn, last changed 2017-09-06 20:09 by christian.heimes. This issue is now closed.

Messages (4)
msg289708 - (view) Author: Suphannee (ssivakorn) Date: 2017-03-16 08:03
1. Allowing attempting to match invalid hostname
According to domain name specification in RFC 1035, only alphanumeric, dot and
hyphen are valid characters in domain name. We observe that
the function match_hostname() in Lib/ allows other special characters (e.g., '=', '&') in hostname when attempting to match with certificate commonName (CN)/subjectAltName DNS. An example would be matching hostname
"" with certificate CN/DNS "" or CN/DNS "*". Ensuring that CN/DNS with invalid characters are rejected, will make the library more robust against attacks that utilize such characters.

2. Matching wildcard in public suffix
As noted in section 7.2 of RFC 6125, some wildcard location specifications are
not clear. We found that the function allows wildcard over public suffix in
certificate as well as allows attempting to match in hostname verification,
e.g., matches hostname "" and "" with
certificate CN/DNS "*.com". This is not an RFC violation, but we might benefit from implementing the check, for example "*.one_label" is restricted. A better option will be having a list of all TLD's and check against it.

msg289709 - (view) Author: Christian Heimes (christian.heimes) * (Python committer) Date: 2017-03-16 08:23
I don't see 1) as a problem. You won't be able to resolve these names in DNS, would you?

Regarding 2). Yes, it would be beneficial to have more elaborate checks to protect against wildcard attacks like *.com. However Python is not a browser. It's really hard to do it right and even harder to keep the rule set up to date. Some TLDs like .uk have sublevel namespaces, e.g. * is also invalid.

The problem is going to shift anyway. For Python 3.7 I'm going to deprecate support for OpenSSL < 1.0.2 and use OpenSSL's hostname verification code instead of ssl.match_hostname().
msg301509 - (view) Author: Alex Gaynor (Alex Gaynor) Date: 2017-09-06 20:06
An additional problem in (2) is that a cert for *.google _is_ legal if the CA can prove that a single organization controls the entire TLD:
msg301510 - (view) Author: Christian Heimes (christian.heimes) * (Python committer) Date: 2017-09-06 20:09
Thanks Alex! Let's get not in the business to verify TLDs.
Date User Action Args
2017-09-06 20:09:50christian.heimessetstatus: open -> closed
resolution: wont fix
messages: + msg301510

stage: resolved
2017-09-06 20:06:37Alex Gaynorsetnosy: + Alex Gaynor
messages: + msg301509
2017-03-16 08:23:41christian.heimessetnosy: + janssen, alex, dstufft
2017-03-16 08:23:23christian.heimessetmessages: + msg289709
2017-03-16 08:03:29ssivakorncreate